MedGemma-27B-IT: benchmark results
Google's MedGemma 27B-IT, an instruction-tuned, image-text-to-text model fine-tuned from Gemma 3 27B for the medical domain. Provider: Google. Released 2025-05-20. Access: Open.
Aggregate Rating 1573 ± 12, rank #1348 of 10916 rated models, from 171 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Oncology Report-Derived VQA - Liver CT Report-Derived | 64.5 | Accuracy (%; zero-shot multiple-choice VQA with 2-5 options | 100 |
| PSF-Med - Open Medical | 9.47 | Paraphrase Flip Rate (%) | 100 |
| PSF-Med - Open Medical - MIMIC-CXR | 6.4 | Paraphrase Flip Rate (%) | 100 |
| PSF-Med - Open Medical - VinDr-CXR | 8.1 | Paraphrase Flip Rate (%) | 100 |
| Open Portuguese LLM - ASSIN2 RTE | 94.56 | Macro F1 (%) | 97 |
| MedLayXPlain | 67.2 | S (self-reported) | 96.8 |
| Open Portuguese LLM - FaQuAD NLI | 83.25 | Macro F1 (%) | 96.4 |
| CURE (Clinical Diagnosis, Physician References) - Top-3 | 85 | Open-ended diagnosis, Hit@3 in percent: share of cases whose | 93.3 |
| CURE (Clinical Diagnosis, RAG) - Top-3 | 28.6 | Open-ended diagnosis, Hit@3 in percent: share of cases whose | 93.3 |
| MI-CXR - Multiple Event Localization | 35.1 | Temporal Event Localization, multiple emergences or resoluti | 92.3 |
| Open Portuguese LLM - ENEM | 78.66 | Accuracy (%) | 92.2 |
| Open Portuguese LLM - OAB Exams | 59.13 | Accuracy (%) | 91.7 |
Interactive version: theaggregate.ai/model?slug=medgemma-27b-it · How It Works · Data refreshed daily, snapshot 2026-11-08.